AI OnDemand (AIoD)
A framework for running bioimage analysis models on any compute environment — local, HPC, or cloud — without requiring users to manage installation or scaling.
Python Nextflow Napari HPC Biology Computer Vision
Overview
AI OnDemand (AIoD) is a framework I developed at the Francis Crick Institute to make ML-based bioimage analysis accessible to wet lab scientists without requiring them to know how to install, configure, or scale models. A researcher can select a model — Cellpose, StarDist, SAM2, and others — and run it locally or submit it to HPC or cloud compute via Nextflow, all from a Napari plugin interface.
The framework separates compute from visualisation: the front-end (Napari) handles data selection and result display; the backend pipeline handles scheduling, containerisation, and scaling.
Key Features
- Any model, any compute — runs on local machines, HPC clusters (via Open OnDemand), or cloud without user-facing configuration changes
- Napari plugin front-end for interactive use and immediate result visualisation
- Nextflow pipeline (Segment-Flow) for scalable, reproducible distributed inference
- Model Registry — Pydantic-based schema and manifests making it straightforward to add new models
- Extensible by design — new models, preprocessing functions, and UIs can be contributed without touching core framework code